Abstract: The detection of the number of space signals is one of the key issues in array signal processing. In view of the poor performance of traditional detection methods at low signal-to-noise ratios, a new method called Detection Technique based on Approximate Eigenvectors (DTAE) is proposed to improve the detection performance of sensor arrays at low signal-to-noise ratios. In the proposed method the direction of the centroid of the cluster of signals is first estimated by some kind of beamform scanning in the space, next the approximate eigenvectors of the data covariance matrix is calculated according to the centroid estimate, then the array output data are weighted by the approximate eigenvectors, finally the estimate of the number of signals is acquired by some kind of manipulating of the peak-to-average power ratio of the weighted data in frequency domain. Simulations show the proposed method DTAE demonstrates much better performance than Akaike Information Criterion (AIC) and other methods at low signal-to-noise ratios, which is valuable in engineering practice.